contextgem

What it is

ContextGem is a free, open-source LLM framework that makes it easier to extract structured data and insights from documents, with minimal code. It supports extracting aspects and concepts, granular reference mapping, and built-in justifications, all within a unified, serializable document model.

How it works

The workflow involves:

• Defining a Document with text (and visuals) to analyze.
• Defining what to extract: Aspects (text segments) and Concepts (data points).
• Running an LLM to perform extraction, optionally using a group of LLMs for complex tasks.

Key components include Document, Aspect, Concept types, and a DocumentLLM helper to run extraction. The README shows examples of adding concepts and aspects, and using llm.extract_all(document).

Getting started

Installation methods shown:

uv add contextgem

or

pip install -U contextgem

Usage example demonstrates creating a Document with raw_text, attaching a StringConcept, configuring a DocumentLLM with a model and API key, and executing extraction via llm.extract_all(doc).

Recent releases

• v0.26.0 (2026-07-28): Added "none" and "max" reasoning effort values, extending accepted set to litellm vocabulary.
• v0.25.1 (2026-06-06): Added license-based dependency audit artifacts (LICENSES.md, licenseal.review.toml) and CI workflow; and fixed Literal type-hint serialization/deserialization.
• v0.25.0 (2026-05-07): Changed concurrency for sub-aspect/sub-concept extraction to use asyncio.gather when use_concurrency=True.
• v0.24.0 (2026-05-05): Added runtime network egress scopes via tethered for non-LLM paths; SaT model downloads handled in _load_sat_model.
• v0.23.0 (2026-05-05): Upgraded pinned dependency versions including litellm, openai, and genai-prices.

Traction

• Stars: 1869

Behind the repo

• License: Apache-2.0
• Language: Python
• Created: 2024-10-11
• Last push: 2026-07-28
• Forks: 160
• Open issues: 0

Caveats

• License: Apache-2.0
• Age: created 2024-10-11; last push 2026-07-28
• Open issues: 0